A Maximum Fitting-based TrAdaBoost Method for Detecting Multiple Subjects' P300 Potentials

Individual difference of brain signal leads to the P300-based interface needs a large amount of training data to construct a pattern recognition model for each subject. Lots of training increases the training cost, and causes the subject's fatigue. TrAdaBoost is a method of transfer a learned classifier's information to another classifier. In order to solve the problem of overfitting caused by combining too many classifiers, a novel maximum fitting-based TrAdaBoost(M-TAB) is proposed to identify the P300 potential across multiple subjects. The M-TAB first trains a classifier with a small number of a subject's data. Then it uses this classifier to adjust the weights of many other classifiers that are trained by other subjects' data. The method retains a high accuracy of 91.05% even if the training data is reduced to 33.33%. The M-TAB improves the accuracy and the information transfer rate by 10.65% and 2.31 bits . min(-1), compared with the traditional training method.